2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
36 static struct vfsmount *shm_mnt;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
69 #include <linux/syscalls.h>
70 #include <linux/fcntl.h>
71 #include <uapi/linux/memfd.h>
73 #include <asm/uaccess.h>
74 #include <asm/pgtable.h>
78 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
79 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
81 /* Pretend that each entry is of this size in directory's i_size */
82 #define BOGO_DIRENT_SIZE 20
84 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
85 #define SHORT_SYMLINK_LEN 128
88 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
89 * inode->i_private (with i_mutex making sure that it has only one user at
90 * a time): we would prefer not to enlarge the shmem inode just for that.
93 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
94 pgoff_t start; /* start of range currently being fallocated */
95 pgoff_t next; /* the next page offset to be fallocated */
96 pgoff_t nr_falloced; /* how many new pages have been fallocated */
97 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
100 /* Flag allocation requirements to shmem_getpage */
102 SGP_READ, /* don't exceed i_size, don't allocate page */
103 SGP_CACHE, /* don't exceed i_size, may allocate page */
104 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
105 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
106 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages / 2;
115 static unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
121 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123 struct shmem_inode_info *info, pgoff_t index);
124 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
127 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
128 struct page **pagep, enum sgp_type sgp, int *fault_type)
130 return shmem_getpage_gfp(inode, index, pagep, sgp,
131 mapping_gfp_mask(inode->i_mapping), fault_type);
134 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
136 return sb->s_fs_info;
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
145 static inline int shmem_acct_size(unsigned long flags, loff_t size)
147 return (flags & VM_NORESERVE) ?
148 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
151 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
153 if (!(flags & VM_NORESERVE))
154 vm_unacct_memory(VM_ACCT(size));
157 static inline int shmem_reacct_size(unsigned long flags,
158 loff_t oldsize, loff_t newsize)
160 if (!(flags & VM_NORESERVE)) {
161 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
162 return security_vm_enough_memory_mm(current->mm,
163 VM_ACCT(newsize) - VM_ACCT(oldsize));
164 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
165 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
171 * ... whereas tmpfs objects are accounted incrementally as
172 * pages are allocated, in order to allow huge sparse files.
173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
176 static inline int shmem_acct_block(unsigned long flags)
178 return (flags & VM_NORESERVE) ?
179 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
182 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
184 if (flags & VM_NORESERVE)
185 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
188 static const struct super_operations shmem_ops;
189 static const struct address_space_operations shmem_aops;
190 static const struct file_operations shmem_file_operations;
191 static const struct inode_operations shmem_inode_operations;
192 static const struct inode_operations shmem_dir_inode_operations;
193 static const struct inode_operations shmem_special_inode_operations;
194 static const struct vm_operations_struct shmem_vm_ops;
196 static LIST_HEAD(shmem_swaplist);
197 static DEFINE_MUTEX(shmem_swaplist_mutex);
199 static int shmem_reserve_inode(struct super_block *sb)
201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202 if (sbinfo->max_inodes) {
203 spin_lock(&sbinfo->stat_lock);
204 if (!sbinfo->free_inodes) {
205 spin_unlock(&sbinfo->stat_lock);
208 sbinfo->free_inodes--;
209 spin_unlock(&sbinfo->stat_lock);
214 static void shmem_free_inode(struct super_block *sb)
216 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
217 if (sbinfo->max_inodes) {
218 spin_lock(&sbinfo->stat_lock);
219 sbinfo->free_inodes++;
220 spin_unlock(&sbinfo->stat_lock);
225 * shmem_recalc_inode - recalculate the block usage of an inode
226 * @inode: inode to recalc
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
234 * It has to be called with the spinlock held.
236 static void shmem_recalc_inode(struct inode *inode)
238 struct shmem_inode_info *info = SHMEM_I(inode);
241 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
243 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
244 if (sbinfo->max_blocks)
245 percpu_counter_add(&sbinfo->used_blocks, -freed);
246 info->alloced -= freed;
247 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
248 shmem_unacct_blocks(info->flags, freed);
253 * Replace item expected in radix tree by a new item, while holding tree lock.
255 static int shmem_radix_tree_replace(struct address_space *mapping,
256 pgoff_t index, void *expected, void *replacement)
261 VM_BUG_ON(!expected);
262 VM_BUG_ON(!replacement);
263 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
266 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
267 if (item != expected)
269 radix_tree_replace_slot(pslot, replacement);
274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
280 static bool shmem_confirm_swap(struct address_space *mapping,
281 pgoff_t index, swp_entry_t swap)
286 item = radix_tree_lookup(&mapping->page_tree, index);
288 return item == swp_to_radix_entry(swap);
292 * Like add_to_page_cache_locked, but error if expected item has gone.
294 static int shmem_add_to_page_cache(struct page *page,
295 struct address_space *mapping,
296 pgoff_t index, void *expected)
300 VM_BUG_ON_PAGE(!PageLocked(page), page);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
303 page_cache_get(page);
304 page->mapping = mapping;
307 spin_lock_irq(&mapping->tree_lock);
309 error = radix_tree_insert(&mapping->page_tree, index, page);
311 error = shmem_radix_tree_replace(mapping, index, expected,
315 __inc_zone_page_state(page, NR_FILE_PAGES);
316 __inc_zone_page_state(page, NR_SHMEM);
317 spin_unlock_irq(&mapping->tree_lock);
319 page->mapping = NULL;
320 spin_unlock_irq(&mapping->tree_lock);
321 page_cache_release(page);
327 * Like delete_from_page_cache, but substitutes swap for page.
329 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
331 struct address_space *mapping = page->mapping;
334 spin_lock_irq(&mapping->tree_lock);
335 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
336 page->mapping = NULL;
338 __dec_zone_page_state(page, NR_FILE_PAGES);
339 __dec_zone_page_state(page, NR_SHMEM);
340 spin_unlock_irq(&mapping->tree_lock);
341 page_cache_release(page);
346 * Remove swap entry from radix tree, free the swap and its page cache.
348 static int shmem_free_swap(struct address_space *mapping,
349 pgoff_t index, void *radswap)
353 spin_lock_irq(&mapping->tree_lock);
354 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
355 spin_unlock_irq(&mapping->tree_lock);
358 free_swap_and_cache(radix_to_swp_entry(radswap));
363 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
365 void shmem_unlock_mapping(struct address_space *mapping)
368 pgoff_t indices[PAGEVEC_SIZE];
371 pagevec_init(&pvec, 0);
373 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
375 while (!mapping_unevictable(mapping)) {
377 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
378 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
380 pvec.nr = find_get_entries(mapping, index,
381 PAGEVEC_SIZE, pvec.pages, indices);
384 index = indices[pvec.nr - 1] + 1;
385 pagevec_remove_exceptionals(&pvec);
386 check_move_unevictable_pages(pvec.pages, pvec.nr);
387 pagevec_release(&pvec);
393 * Remove range of pages and swap entries from radix tree, and free them.
394 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
396 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
399 struct address_space *mapping = inode->i_mapping;
400 struct shmem_inode_info *info = SHMEM_I(inode);
401 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
402 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
403 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
404 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
406 pgoff_t indices[PAGEVEC_SIZE];
407 long nr_swaps_freed = 0;
412 end = -1; /* unsigned, so actually very big */
414 pagevec_init(&pvec, 0);
416 while (index < end) {
417 pvec.nr = find_get_entries(mapping, index,
418 min(end - index, (pgoff_t)PAGEVEC_SIZE),
419 pvec.pages, indices);
422 for (i = 0; i < pagevec_count(&pvec); i++) {
423 struct page *page = pvec.pages[i];
429 if (radix_tree_exceptional_entry(page)) {
432 nr_swaps_freed += !shmem_free_swap(mapping,
437 if (!trylock_page(page))
439 if (!unfalloc || !PageUptodate(page)) {
440 if (page->mapping == mapping) {
441 VM_BUG_ON_PAGE(PageWriteback(page), page);
442 truncate_inode_page(mapping, page);
447 pagevec_remove_exceptionals(&pvec);
448 pagevec_release(&pvec);
454 struct page *page = NULL;
455 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
457 unsigned int top = PAGE_CACHE_SIZE;
462 zero_user_segment(page, partial_start, top);
463 set_page_dirty(page);
465 page_cache_release(page);
469 struct page *page = NULL;
470 shmem_getpage(inode, end, &page, SGP_READ, NULL);
472 zero_user_segment(page, 0, partial_end);
473 set_page_dirty(page);
475 page_cache_release(page);
482 while (index < end) {
485 pvec.nr = find_get_entries(mapping, index,
486 min(end - index, (pgoff_t)PAGEVEC_SIZE),
487 pvec.pages, indices);
489 /* If all gone or hole-punch or unfalloc, we're done */
490 if (index == start || end != -1)
492 /* But if truncating, restart to make sure all gone */
496 for (i = 0; i < pagevec_count(&pvec); i++) {
497 struct page *page = pvec.pages[i];
503 if (radix_tree_exceptional_entry(page)) {
506 if (shmem_free_swap(mapping, index, page)) {
507 /* Swap was replaced by page: retry */
516 if (!unfalloc || !PageUptodate(page)) {
517 if (page->mapping == mapping) {
518 VM_BUG_ON_PAGE(PageWriteback(page), page);
519 truncate_inode_page(mapping, page);
521 /* Page was replaced by swap: retry */
529 pagevec_remove_exceptionals(&pvec);
530 pagevec_release(&pvec);
534 spin_lock(&info->lock);
535 info->swapped -= nr_swaps_freed;
536 shmem_recalc_inode(inode);
537 spin_unlock(&info->lock);
540 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
542 shmem_undo_range(inode, lstart, lend, false);
543 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
545 EXPORT_SYMBOL_GPL(shmem_truncate_range);
547 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
550 struct inode *inode = dentry->d_inode;
551 struct shmem_inode_info *info = SHMEM_I(inode);
553 spin_lock(&info->lock);
554 shmem_recalc_inode(inode);
555 spin_unlock(&info->lock);
557 generic_fillattr(inode, stat);
562 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
564 struct inode *inode = d_inode(dentry);
565 struct shmem_inode_info *info = SHMEM_I(inode);
568 error = inode_change_ok(inode, attr);
572 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
573 loff_t oldsize = inode->i_size;
574 loff_t newsize = attr->ia_size;
576 /* protected by i_mutex */
577 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
578 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
581 if (newsize != oldsize) {
582 error = shmem_reacct_size(SHMEM_I(inode)->flags,
586 i_size_write(inode, newsize);
587 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
589 if (newsize <= oldsize) {
590 loff_t holebegin = round_up(newsize, PAGE_SIZE);
591 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
592 shmem_truncate_range(inode, newsize, (loff_t)-1);
593 /* unmap again to remove racily COWed private pages */
594 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
598 setattr_copy(inode, attr);
599 if (attr->ia_valid & ATTR_MODE)
600 error = posix_acl_chmod(inode, inode->i_mode);
604 static void shmem_evict_inode(struct inode *inode)
606 struct shmem_inode_info *info = SHMEM_I(inode);
608 if (inode->i_mapping->a_ops == &shmem_aops) {
609 shmem_unacct_size(info->flags, inode->i_size);
611 shmem_truncate_range(inode, 0, (loff_t)-1);
612 if (!list_empty(&info->swaplist)) {
613 mutex_lock(&shmem_swaplist_mutex);
614 list_del_init(&info->swaplist);
615 mutex_unlock(&shmem_swaplist_mutex);
618 kfree(info->symlink);
620 simple_xattrs_free(&info->xattrs);
621 WARN_ON(inode->i_blocks);
622 shmem_free_inode(inode->i_sb);
627 * If swap found in inode, free it and move page from swapcache to filecache.
629 static int shmem_unuse_inode(struct shmem_inode_info *info,
630 swp_entry_t swap, struct page **pagep)
632 struct address_space *mapping = info->vfs_inode.i_mapping;
638 radswap = swp_to_radix_entry(swap);
639 index = radix_tree_locate_item(&mapping->page_tree, radswap);
641 return -EAGAIN; /* tell shmem_unuse we found nothing */
644 * Move _head_ to start search for next from here.
645 * But be careful: shmem_evict_inode checks list_empty without taking
646 * mutex, and there's an instant in list_move_tail when info->swaplist
647 * would appear empty, if it were the only one on shmem_swaplist.
649 if (shmem_swaplist.next != &info->swaplist)
650 list_move_tail(&shmem_swaplist, &info->swaplist);
652 gfp = mapping_gfp_mask(mapping);
653 if (shmem_should_replace_page(*pagep, gfp)) {
654 mutex_unlock(&shmem_swaplist_mutex);
655 error = shmem_replace_page(pagep, gfp, info, index);
656 mutex_lock(&shmem_swaplist_mutex);
658 * We needed to drop mutex to make that restrictive page
659 * allocation, but the inode might have been freed while we
660 * dropped it: although a racing shmem_evict_inode() cannot
661 * complete without emptying the radix_tree, our page lock
662 * on this swapcache page is not enough to prevent that -
663 * free_swap_and_cache() of our swap entry will only
664 * trylock_page(), removing swap from radix_tree whatever.
666 * We must not proceed to shmem_add_to_page_cache() if the
667 * inode has been freed, but of course we cannot rely on
668 * inode or mapping or info to check that. However, we can
669 * safely check if our swap entry is still in use (and here
670 * it can't have got reused for another page): if it's still
671 * in use, then the inode cannot have been freed yet, and we
672 * can safely proceed (if it's no longer in use, that tells
673 * nothing about the inode, but we don't need to unuse swap).
675 if (!page_swapcount(*pagep))
680 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
681 * but also to hold up shmem_evict_inode(): so inode cannot be freed
682 * beneath us (pagelock doesn't help until the page is in pagecache).
685 error = shmem_add_to_page_cache(*pagep, mapping, index,
687 if (error != -ENOMEM) {
689 * Truncation and eviction use free_swap_and_cache(), which
690 * only does trylock page: if we raced, best clean up here.
692 delete_from_swap_cache(*pagep);
693 set_page_dirty(*pagep);
695 spin_lock(&info->lock);
697 spin_unlock(&info->lock);
705 * Search through swapped inodes to find and replace swap by page.
707 int shmem_unuse(swp_entry_t swap, struct page *page)
709 struct list_head *this, *next;
710 struct shmem_inode_info *info;
711 struct mem_cgroup *memcg;
715 * There's a faint possibility that swap page was replaced before
716 * caller locked it: caller will come back later with the right page.
718 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
722 * Charge page using GFP_KERNEL while we can wait, before taking
723 * the shmem_swaplist_mutex which might hold up shmem_writepage().
724 * Charged back to the user (not to caller) when swap account is used.
726 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
730 /* No radix_tree_preload: swap entry keeps a place for page in tree */
733 mutex_lock(&shmem_swaplist_mutex);
734 list_for_each_safe(this, next, &shmem_swaplist) {
735 info = list_entry(this, struct shmem_inode_info, swaplist);
737 error = shmem_unuse_inode(info, swap, &page);
739 list_del_init(&info->swaplist);
741 if (error != -EAGAIN)
743 /* found nothing in this: move on to search the next */
745 mutex_unlock(&shmem_swaplist_mutex);
748 if (error != -ENOMEM)
750 mem_cgroup_cancel_charge(page, memcg, false);
752 mem_cgroup_commit_charge(page, memcg, true, false);
755 page_cache_release(page);
760 * Move the page from the page cache to the swap cache.
762 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
764 struct shmem_inode_info *info;
765 struct address_space *mapping;
770 BUG_ON(!PageLocked(page));
771 mapping = page->mapping;
773 inode = mapping->host;
774 info = SHMEM_I(inode);
775 if (info->flags & VM_LOCKED)
777 if (!total_swap_pages)
781 * Our capabilities prevent regular writeback or sync from ever calling
782 * shmem_writepage; but a stacking filesystem might use ->writepage of
783 * its underlying filesystem, in which case tmpfs should write out to
784 * swap only in response to memory pressure, and not for the writeback
787 if (!wbc->for_reclaim) {
788 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
793 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
794 * value into swapfile.c, the only way we can correctly account for a
795 * fallocated page arriving here is now to initialize it and write it.
797 * That's okay for a page already fallocated earlier, but if we have
798 * not yet completed the fallocation, then (a) we want to keep track
799 * of this page in case we have to undo it, and (b) it may not be a
800 * good idea to continue anyway, once we're pushing into swap. So
801 * reactivate the page, and let shmem_fallocate() quit when too many.
803 if (!PageUptodate(page)) {
804 if (inode->i_private) {
805 struct shmem_falloc *shmem_falloc;
806 spin_lock(&inode->i_lock);
807 shmem_falloc = inode->i_private;
809 !shmem_falloc->waitq &&
810 index >= shmem_falloc->start &&
811 index < shmem_falloc->next)
812 shmem_falloc->nr_unswapped++;
815 spin_unlock(&inode->i_lock);
819 clear_highpage(page);
820 flush_dcache_page(page);
821 SetPageUptodate(page);
824 swap = get_swap_page();
829 * Add inode to shmem_unuse()'s list of swapped-out inodes,
830 * if it's not already there. Do it now before the page is
831 * moved to swap cache, when its pagelock no longer protects
832 * the inode from eviction. But don't unlock the mutex until
833 * we've incremented swapped, because shmem_unuse_inode() will
834 * prune a !swapped inode from the swaplist under this mutex.
836 mutex_lock(&shmem_swaplist_mutex);
837 if (list_empty(&info->swaplist))
838 list_add_tail(&info->swaplist, &shmem_swaplist);
840 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
841 swap_shmem_alloc(swap);
842 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
844 spin_lock(&info->lock);
846 shmem_recalc_inode(inode);
847 spin_unlock(&info->lock);
849 mutex_unlock(&shmem_swaplist_mutex);
850 BUG_ON(page_mapped(page));
851 swap_writepage(page, wbc);
855 mutex_unlock(&shmem_swaplist_mutex);
856 swapcache_free(swap);
858 set_page_dirty(page);
859 if (wbc->for_reclaim)
860 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
867 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
871 if (!mpol || mpol->mode == MPOL_DEFAULT)
872 return; /* show nothing */
874 mpol_to_str(buffer, sizeof(buffer), mpol);
876 seq_printf(seq, ",mpol=%s", buffer);
879 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
881 struct mempolicy *mpol = NULL;
883 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
886 spin_unlock(&sbinfo->stat_lock);
890 #endif /* CONFIG_TMPFS */
892 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
893 struct shmem_inode_info *info, pgoff_t index)
895 struct vm_area_struct pvma;
898 /* Create a pseudo vma that just contains the policy */
900 /* Bias interleave by inode number to distribute better across nodes */
901 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
903 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
905 page = swapin_readahead(swap, gfp, &pvma, 0);
907 /* Drop reference taken by mpol_shared_policy_lookup() */
908 mpol_cond_put(pvma.vm_policy);
913 static struct page *shmem_alloc_page(gfp_t gfp,
914 struct shmem_inode_info *info, pgoff_t index)
916 struct vm_area_struct pvma;
919 /* Create a pseudo vma that just contains the policy */
921 /* Bias interleave by inode number to distribute better across nodes */
922 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
924 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
926 page = alloc_page_vma(gfp, &pvma, 0);
928 /* Drop reference taken by mpol_shared_policy_lookup() */
929 mpol_cond_put(pvma.vm_policy);
933 #else /* !CONFIG_NUMA */
935 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
938 #endif /* CONFIG_TMPFS */
940 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
941 struct shmem_inode_info *info, pgoff_t index)
943 return swapin_readahead(swap, gfp, NULL, 0);
946 static inline struct page *shmem_alloc_page(gfp_t gfp,
947 struct shmem_inode_info *info, pgoff_t index)
949 return alloc_page(gfp);
951 #endif /* CONFIG_NUMA */
953 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
954 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
961 * When a page is moved from swapcache to shmem filecache (either by the
962 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
963 * shmem_unuse_inode()), it may have been read in earlier from swap, in
964 * ignorance of the mapping it belongs to. If that mapping has special
965 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
966 * we may need to copy to a suitable page before moving to filecache.
968 * In a future release, this may well be extended to respect cpuset and
969 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
970 * but for now it is a simple matter of zone.
972 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
974 return page_zonenum(page) > gfp_zone(gfp);
977 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
978 struct shmem_inode_info *info, pgoff_t index)
980 struct page *oldpage, *newpage;
981 struct address_space *swap_mapping;
986 swap_index = page_private(oldpage);
987 swap_mapping = page_mapping(oldpage);
990 * We have arrived here because our zones are constrained, so don't
991 * limit chance of success by further cpuset and node constraints.
993 gfp &= ~GFP_CONSTRAINT_MASK;
994 newpage = shmem_alloc_page(gfp, info, index);
998 page_cache_get(newpage);
999 copy_highpage(newpage, oldpage);
1000 flush_dcache_page(newpage);
1002 __SetPageLocked(newpage);
1003 SetPageUptodate(newpage);
1004 SetPageSwapBacked(newpage);
1005 set_page_private(newpage, swap_index);
1006 SetPageSwapCache(newpage);
1009 * Our caller will very soon move newpage out of swapcache, but it's
1010 * a nice clean interface for us to replace oldpage by newpage there.
1012 spin_lock_irq(&swap_mapping->tree_lock);
1013 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1016 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1017 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1019 spin_unlock_irq(&swap_mapping->tree_lock);
1021 if (unlikely(error)) {
1023 * Is this possible? I think not, now that our callers check
1024 * both PageSwapCache and page_private after getting page lock;
1025 * but be defensive. Reverse old to newpage for clear and free.
1029 mem_cgroup_replace_page(oldpage, newpage);
1030 lru_cache_add_anon(newpage);
1034 ClearPageSwapCache(oldpage);
1035 set_page_private(oldpage, 0);
1037 unlock_page(oldpage);
1038 page_cache_release(oldpage);
1039 page_cache_release(oldpage);
1044 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1046 * If we allocate a new one we do not mark it dirty. That's up to the
1047 * vm. If we swap it in we mark it dirty since we also free the swap
1048 * entry since a page cannot live in both the swap and page cache
1050 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1051 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1053 struct address_space *mapping = inode->i_mapping;
1054 struct shmem_inode_info *info;
1055 struct shmem_sb_info *sbinfo;
1056 struct mem_cgroup *memcg;
1063 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1067 page = find_lock_entry(mapping, index);
1068 if (radix_tree_exceptional_entry(page)) {
1069 swap = radix_to_swp_entry(page);
1073 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1074 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1079 if (page && sgp == SGP_WRITE)
1080 mark_page_accessed(page);
1082 /* fallocated page? */
1083 if (page && !PageUptodate(page)) {
1084 if (sgp != SGP_READ)
1087 page_cache_release(page);
1090 if (page || (sgp == SGP_READ && !swap.val)) {
1096 * Fast cache lookup did not find it:
1097 * bring it back from swap or allocate.
1099 info = SHMEM_I(inode);
1100 sbinfo = SHMEM_SB(inode->i_sb);
1103 /* Look it up and read it in.. */
1104 page = lookup_swap_cache(swap);
1106 /* here we actually do the io */
1108 *fault_type |= VM_FAULT_MAJOR;
1109 page = shmem_swapin(swap, gfp, info, index);
1116 /* We have to do this with page locked to prevent races */
1118 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1119 !shmem_confirm_swap(mapping, index, swap)) {
1120 error = -EEXIST; /* try again */
1123 if (!PageUptodate(page)) {
1127 wait_on_page_writeback(page);
1129 if (shmem_should_replace_page(page, gfp)) {
1130 error = shmem_replace_page(&page, gfp, info, index);
1135 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1138 error = shmem_add_to_page_cache(page, mapping, index,
1139 swp_to_radix_entry(swap));
1141 * We already confirmed swap under page lock, and make
1142 * no memory allocation here, so usually no possibility
1143 * of error; but free_swap_and_cache() only trylocks a
1144 * page, so it is just possible that the entry has been
1145 * truncated or holepunched since swap was confirmed.
1146 * shmem_undo_range() will have done some of the
1147 * unaccounting, now delete_from_swap_cache() will do
1149 * Reset swap.val? No, leave it so "failed" goes back to
1150 * "repeat": reading a hole and writing should succeed.
1153 mem_cgroup_cancel_charge(page, memcg, false);
1154 delete_from_swap_cache(page);
1160 mem_cgroup_commit_charge(page, memcg, true, false);
1162 spin_lock(&info->lock);
1164 shmem_recalc_inode(inode);
1165 spin_unlock(&info->lock);
1167 if (sgp == SGP_WRITE)
1168 mark_page_accessed(page);
1170 delete_from_swap_cache(page);
1171 set_page_dirty(page);
1175 if (shmem_acct_block(info->flags)) {
1179 if (sbinfo->max_blocks) {
1180 if (percpu_counter_compare(&sbinfo->used_blocks,
1181 sbinfo->max_blocks) >= 0) {
1185 percpu_counter_inc(&sbinfo->used_blocks);
1188 page = shmem_alloc_page(gfp, info, index);
1194 __SetPageSwapBacked(page);
1195 __SetPageLocked(page);
1196 if (sgp == SGP_WRITE)
1197 __SetPageReferenced(page);
1199 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1203 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1205 error = shmem_add_to_page_cache(page, mapping, index,
1207 radix_tree_preload_end();
1210 mem_cgroup_cancel_charge(page, memcg, false);
1213 mem_cgroup_commit_charge(page, memcg, false, false);
1214 lru_cache_add_anon(page);
1216 spin_lock(&info->lock);
1218 inode->i_blocks += BLOCKS_PER_PAGE;
1219 shmem_recalc_inode(inode);
1220 spin_unlock(&info->lock);
1224 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1226 if (sgp == SGP_FALLOC)
1230 * Let SGP_WRITE caller clear ends if write does not fill page;
1231 * but SGP_FALLOC on a page fallocated earlier must initialize
1232 * it now, lest undo on failure cancel our earlier guarantee.
1234 if (sgp != SGP_WRITE) {
1235 clear_highpage(page);
1236 flush_dcache_page(page);
1237 SetPageUptodate(page);
1239 if (sgp == SGP_DIRTY)
1240 set_page_dirty(page);
1243 /* Perhaps the file has been truncated since we checked */
1244 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1245 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1259 info = SHMEM_I(inode);
1260 ClearPageDirty(page);
1261 delete_from_page_cache(page);
1262 spin_lock(&info->lock);
1264 inode->i_blocks -= BLOCKS_PER_PAGE;
1265 spin_unlock(&info->lock);
1267 sbinfo = SHMEM_SB(inode->i_sb);
1268 if (sbinfo->max_blocks)
1269 percpu_counter_add(&sbinfo->used_blocks, -1);
1271 shmem_unacct_blocks(info->flags, 1);
1273 if (swap.val && error != -EINVAL &&
1274 !shmem_confirm_swap(mapping, index, swap))
1279 page_cache_release(page);
1281 if (error == -ENOSPC && !once++) {
1282 info = SHMEM_I(inode);
1283 spin_lock(&info->lock);
1284 shmem_recalc_inode(inode);
1285 spin_unlock(&info->lock);
1288 if (error == -EEXIST) /* from above or from radix_tree_insert */
1293 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1295 struct inode *inode = file_inode(vma->vm_file);
1297 int ret = VM_FAULT_LOCKED;
1300 * Trinity finds that probing a hole which tmpfs is punching can
1301 * prevent the hole-punch from ever completing: which in turn
1302 * locks writers out with its hold on i_mutex. So refrain from
1303 * faulting pages into the hole while it's being punched. Although
1304 * shmem_undo_range() does remove the additions, it may be unable to
1305 * keep up, as each new page needs its own unmap_mapping_range() call,
1306 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1308 * It does not matter if we sometimes reach this check just before the
1309 * hole-punch begins, so that one fault then races with the punch:
1310 * we just need to make racing faults a rare case.
1312 * The implementation below would be much simpler if we just used a
1313 * standard mutex or completion: but we cannot take i_mutex in fault,
1314 * and bloating every shmem inode for this unlikely case would be sad.
1316 if (unlikely(inode->i_private)) {
1317 struct shmem_falloc *shmem_falloc;
1319 spin_lock(&inode->i_lock);
1320 shmem_falloc = inode->i_private;
1322 shmem_falloc->waitq &&
1323 vmf->pgoff >= shmem_falloc->start &&
1324 vmf->pgoff < shmem_falloc->next) {
1325 wait_queue_head_t *shmem_falloc_waitq;
1326 DEFINE_WAIT(shmem_fault_wait);
1328 ret = VM_FAULT_NOPAGE;
1329 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1330 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1331 /* It's polite to up mmap_sem if we can */
1332 up_read(&vma->vm_mm->mmap_sem);
1333 ret = VM_FAULT_RETRY;
1336 shmem_falloc_waitq = shmem_falloc->waitq;
1337 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1338 TASK_UNINTERRUPTIBLE);
1339 spin_unlock(&inode->i_lock);
1343 * shmem_falloc_waitq points into the shmem_fallocate()
1344 * stack of the hole-punching task: shmem_falloc_waitq
1345 * is usually invalid by the time we reach here, but
1346 * finish_wait() does not dereference it in that case;
1347 * though i_lock needed lest racing with wake_up_all().
1349 spin_lock(&inode->i_lock);
1350 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1351 spin_unlock(&inode->i_lock);
1354 spin_unlock(&inode->i_lock);
1357 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1359 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1361 if (ret & VM_FAULT_MAJOR) {
1362 count_vm_event(PGMAJFAULT);
1363 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1369 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1371 struct inode *inode = file_inode(vma->vm_file);
1372 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1375 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1378 struct inode *inode = file_inode(vma->vm_file);
1381 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1382 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1386 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1388 struct inode *inode = file_inode(file);
1389 struct shmem_inode_info *info = SHMEM_I(inode);
1390 int retval = -ENOMEM;
1392 spin_lock(&info->lock);
1393 if (lock && !(info->flags & VM_LOCKED)) {
1394 if (!user_shm_lock(inode->i_size, user))
1396 info->flags |= VM_LOCKED;
1397 mapping_set_unevictable(file->f_mapping);
1399 if (!lock && (info->flags & VM_LOCKED) && user) {
1400 user_shm_unlock(inode->i_size, user);
1401 info->flags &= ~VM_LOCKED;
1402 mapping_clear_unevictable(file->f_mapping);
1407 spin_unlock(&info->lock);
1411 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1413 file_accessed(file);
1414 vma->vm_ops = &shmem_vm_ops;
1418 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1419 umode_t mode, dev_t dev, unsigned long flags)
1421 struct inode *inode;
1422 struct shmem_inode_info *info;
1423 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1425 if (shmem_reserve_inode(sb))
1428 inode = new_inode(sb);
1430 inode->i_ino = get_next_ino();
1431 inode_init_owner(inode, dir, mode);
1432 inode->i_blocks = 0;
1433 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1434 inode->i_generation = get_seconds();
1435 info = SHMEM_I(inode);
1436 memset(info, 0, (char *)inode - (char *)info);
1437 spin_lock_init(&info->lock);
1438 info->seals = F_SEAL_SEAL;
1439 info->flags = flags & VM_NORESERVE;
1440 INIT_LIST_HEAD(&info->swaplist);
1441 simple_xattrs_init(&info->xattrs);
1442 cache_no_acl(inode);
1444 switch (mode & S_IFMT) {
1446 inode->i_op = &shmem_special_inode_operations;
1447 init_special_inode(inode, mode, dev);
1450 inode->i_mapping->a_ops = &shmem_aops;
1451 inode->i_op = &shmem_inode_operations;
1452 inode->i_fop = &shmem_file_operations;
1453 mpol_shared_policy_init(&info->policy,
1454 shmem_get_sbmpol(sbinfo));
1458 /* Some things misbehave if size == 0 on a directory */
1459 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1460 inode->i_op = &shmem_dir_inode_operations;
1461 inode->i_fop = &simple_dir_operations;
1465 * Must not load anything in the rbtree,
1466 * mpol_free_shared_policy will not be called.
1468 mpol_shared_policy_init(&info->policy, NULL);
1472 shmem_free_inode(sb);
1476 bool shmem_mapping(struct address_space *mapping)
1481 return mapping->host->i_sb->s_op == &shmem_ops;
1485 static const struct inode_operations shmem_symlink_inode_operations;
1486 static const struct inode_operations shmem_short_symlink_operations;
1488 #ifdef CONFIG_TMPFS_XATTR
1489 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1491 #define shmem_initxattrs NULL
1495 shmem_write_begin(struct file *file, struct address_space *mapping,
1496 loff_t pos, unsigned len, unsigned flags,
1497 struct page **pagep, void **fsdata)
1499 struct inode *inode = mapping->host;
1500 struct shmem_inode_info *info = SHMEM_I(inode);
1501 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1503 /* i_mutex is held by caller */
1504 if (unlikely(info->seals)) {
1505 if (info->seals & F_SEAL_WRITE)
1507 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1511 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1515 shmem_write_end(struct file *file, struct address_space *mapping,
1516 loff_t pos, unsigned len, unsigned copied,
1517 struct page *page, void *fsdata)
1519 struct inode *inode = mapping->host;
1521 if (pos + copied > inode->i_size)
1522 i_size_write(inode, pos + copied);
1524 if (!PageUptodate(page)) {
1525 if (copied < PAGE_CACHE_SIZE) {
1526 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1527 zero_user_segments(page, 0, from,
1528 from + copied, PAGE_CACHE_SIZE);
1530 SetPageUptodate(page);
1532 set_page_dirty(page);
1534 page_cache_release(page);
1539 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1541 struct file *file = iocb->ki_filp;
1542 struct inode *inode = file_inode(file);
1543 struct address_space *mapping = inode->i_mapping;
1545 unsigned long offset;
1546 enum sgp_type sgp = SGP_READ;
1549 loff_t *ppos = &iocb->ki_pos;
1552 * Might this read be for a stacking filesystem? Then when reading
1553 * holes of a sparse file, we actually need to allocate those pages,
1554 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1556 if (!iter_is_iovec(to))
1559 index = *ppos >> PAGE_CACHE_SHIFT;
1560 offset = *ppos & ~PAGE_CACHE_MASK;
1563 struct page *page = NULL;
1565 unsigned long nr, ret;
1566 loff_t i_size = i_size_read(inode);
1568 end_index = i_size >> PAGE_CACHE_SHIFT;
1569 if (index > end_index)
1571 if (index == end_index) {
1572 nr = i_size & ~PAGE_CACHE_MASK;
1577 error = shmem_getpage(inode, index, &page, sgp, NULL);
1579 if (error == -EINVAL)
1587 * We must evaluate after, since reads (unlike writes)
1588 * are called without i_mutex protection against truncate
1590 nr = PAGE_CACHE_SIZE;
1591 i_size = i_size_read(inode);
1592 end_index = i_size >> PAGE_CACHE_SHIFT;
1593 if (index == end_index) {
1594 nr = i_size & ~PAGE_CACHE_MASK;
1597 page_cache_release(page);
1605 * If users can be writing to this page using arbitrary
1606 * virtual addresses, take care about potential aliasing
1607 * before reading the page on the kernel side.
1609 if (mapping_writably_mapped(mapping))
1610 flush_dcache_page(page);
1612 * Mark the page accessed if we read the beginning.
1615 mark_page_accessed(page);
1617 page = ZERO_PAGE(0);
1618 page_cache_get(page);
1622 * Ok, we have the page, and it's up-to-date, so
1623 * now we can copy it to user space...
1625 ret = copy_page_to_iter(page, offset, nr, to);
1628 index += offset >> PAGE_CACHE_SHIFT;
1629 offset &= ~PAGE_CACHE_MASK;
1631 page_cache_release(page);
1632 if (!iov_iter_count(to))
1641 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1642 file_accessed(file);
1643 return retval ? retval : error;
1646 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1647 struct pipe_inode_info *pipe, size_t len,
1650 struct address_space *mapping = in->f_mapping;
1651 struct inode *inode = mapping->host;
1652 unsigned int loff, nr_pages, req_pages;
1653 struct page *pages[PIPE_DEF_BUFFERS];
1654 struct partial_page partial[PIPE_DEF_BUFFERS];
1656 pgoff_t index, end_index;
1659 struct splice_pipe_desc spd = {
1662 .nr_pages_max = PIPE_DEF_BUFFERS,
1664 .ops = &page_cache_pipe_buf_ops,
1665 .spd_release = spd_release_page,
1668 isize = i_size_read(inode);
1669 if (unlikely(*ppos >= isize))
1672 left = isize - *ppos;
1673 if (unlikely(left < len))
1676 if (splice_grow_spd(pipe, &spd))
1679 index = *ppos >> PAGE_CACHE_SHIFT;
1680 loff = *ppos & ~PAGE_CACHE_MASK;
1681 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1682 nr_pages = min(req_pages, spd.nr_pages_max);
1684 spd.nr_pages = find_get_pages_contig(mapping, index,
1685 nr_pages, spd.pages);
1686 index += spd.nr_pages;
1689 while (spd.nr_pages < nr_pages) {
1690 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1694 spd.pages[spd.nr_pages++] = page;
1698 index = *ppos >> PAGE_CACHE_SHIFT;
1699 nr_pages = spd.nr_pages;
1702 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1703 unsigned int this_len;
1708 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1709 page = spd.pages[page_nr];
1711 if (!PageUptodate(page) || page->mapping != mapping) {
1712 error = shmem_getpage(inode, index, &page,
1717 page_cache_release(spd.pages[page_nr]);
1718 spd.pages[page_nr] = page;
1721 isize = i_size_read(inode);
1722 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1723 if (unlikely(!isize || index > end_index))
1726 if (end_index == index) {
1729 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1733 this_len = min(this_len, plen - loff);
1737 spd.partial[page_nr].offset = loff;
1738 spd.partial[page_nr].len = this_len;
1745 while (page_nr < nr_pages)
1746 page_cache_release(spd.pages[page_nr++]);
1749 error = splice_to_pipe(pipe, &spd);
1751 splice_shrink_spd(&spd);
1761 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1763 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1764 pgoff_t index, pgoff_t end, int whence)
1767 struct pagevec pvec;
1768 pgoff_t indices[PAGEVEC_SIZE];
1772 pagevec_init(&pvec, 0);
1773 pvec.nr = 1; /* start small: we may be there already */
1775 pvec.nr = find_get_entries(mapping, index,
1776 pvec.nr, pvec.pages, indices);
1778 if (whence == SEEK_DATA)
1782 for (i = 0; i < pvec.nr; i++, index++) {
1783 if (index < indices[i]) {
1784 if (whence == SEEK_HOLE) {
1790 page = pvec.pages[i];
1791 if (page && !radix_tree_exceptional_entry(page)) {
1792 if (!PageUptodate(page))
1796 (page && whence == SEEK_DATA) ||
1797 (!page && whence == SEEK_HOLE)) {
1802 pagevec_remove_exceptionals(&pvec);
1803 pagevec_release(&pvec);
1804 pvec.nr = PAGEVEC_SIZE;
1810 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1812 struct address_space *mapping = file->f_mapping;
1813 struct inode *inode = mapping->host;
1817 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1818 return generic_file_llseek_size(file, offset, whence,
1819 MAX_LFS_FILESIZE, i_size_read(inode));
1820 mutex_lock(&inode->i_mutex);
1821 /* We're holding i_mutex so we can access i_size directly */
1825 else if (offset >= inode->i_size)
1828 start = offset >> PAGE_CACHE_SHIFT;
1829 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1830 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1831 new_offset <<= PAGE_CACHE_SHIFT;
1832 if (new_offset > offset) {
1833 if (new_offset < inode->i_size)
1834 offset = new_offset;
1835 else if (whence == SEEK_DATA)
1838 offset = inode->i_size;
1843 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1844 mutex_unlock(&inode->i_mutex);
1849 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1850 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1852 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1853 #define LAST_SCAN 4 /* about 150ms max */
1855 static void shmem_tag_pins(struct address_space *mapping)
1857 struct radix_tree_iter iter;
1867 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1868 page = radix_tree_deref_slot(slot);
1869 if (!page || radix_tree_exception(page)) {
1870 if (radix_tree_deref_retry(page))
1872 } else if (page_count(page) - page_mapcount(page) > 1) {
1873 spin_lock_irq(&mapping->tree_lock);
1874 radix_tree_tag_set(&mapping->page_tree, iter.index,
1876 spin_unlock_irq(&mapping->tree_lock);
1879 if (need_resched()) {
1881 start = iter.index + 1;
1889 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1890 * via get_user_pages(), drivers might have some pending I/O without any active
1891 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1892 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1893 * them to be dropped.
1894 * The caller must guarantee that no new user will acquire writable references
1895 * to those pages to avoid races.
1897 static int shmem_wait_for_pins(struct address_space *mapping)
1899 struct radix_tree_iter iter;
1905 shmem_tag_pins(mapping);
1908 for (scan = 0; scan <= LAST_SCAN; scan++) {
1909 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1913 lru_add_drain_all();
1914 else if (schedule_timeout_killable((HZ << scan) / 200))
1920 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
1921 start, SHMEM_TAG_PINNED) {
1923 page = radix_tree_deref_slot(slot);
1924 if (radix_tree_exception(page)) {
1925 if (radix_tree_deref_retry(page))
1932 page_count(page) - page_mapcount(page) != 1) {
1933 if (scan < LAST_SCAN)
1934 goto continue_resched;
1937 * On the last scan, we clean up all those tags
1938 * we inserted; but make a note that we still
1939 * found pages pinned.
1944 spin_lock_irq(&mapping->tree_lock);
1945 radix_tree_tag_clear(&mapping->page_tree,
1946 iter.index, SHMEM_TAG_PINNED);
1947 spin_unlock_irq(&mapping->tree_lock);
1949 if (need_resched()) {
1951 start = iter.index + 1;
1961 #define F_ALL_SEALS (F_SEAL_SEAL | \
1966 int shmem_add_seals(struct file *file, unsigned int seals)
1968 struct inode *inode = file_inode(file);
1969 struct shmem_inode_info *info = SHMEM_I(inode);
1974 * Sealing allows multiple parties to share a shmem-file but restrict
1975 * access to a specific subset of file operations. Seals can only be
1976 * added, but never removed. This way, mutually untrusted parties can
1977 * share common memory regions with a well-defined policy. A malicious
1978 * peer can thus never perform unwanted operations on a shared object.
1980 * Seals are only supported on special shmem-files and always affect
1981 * the whole underlying inode. Once a seal is set, it may prevent some
1982 * kinds of access to the file. Currently, the following seals are
1984 * SEAL_SEAL: Prevent further seals from being set on this file
1985 * SEAL_SHRINK: Prevent the file from shrinking
1986 * SEAL_GROW: Prevent the file from growing
1987 * SEAL_WRITE: Prevent write access to the file
1989 * As we don't require any trust relationship between two parties, we
1990 * must prevent seals from being removed. Therefore, sealing a file
1991 * only adds a given set of seals to the file, it never touches
1992 * existing seals. Furthermore, the "setting seals"-operation can be
1993 * sealed itself, which basically prevents any further seal from being
1996 * Semantics of sealing are only defined on volatile files. Only
1997 * anonymous shmem files support sealing. More importantly, seals are
1998 * never written to disk. Therefore, there's no plan to support it on
2002 if (file->f_op != &shmem_file_operations)
2004 if (!(file->f_mode & FMODE_WRITE))
2006 if (seals & ~(unsigned int)F_ALL_SEALS)
2009 mutex_lock(&inode->i_mutex);
2011 if (info->seals & F_SEAL_SEAL) {
2016 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2017 error = mapping_deny_writable(file->f_mapping);
2021 error = shmem_wait_for_pins(file->f_mapping);
2023 mapping_allow_writable(file->f_mapping);
2028 info->seals |= seals;
2032 mutex_unlock(&inode->i_mutex);
2035 EXPORT_SYMBOL_GPL(shmem_add_seals);
2037 int shmem_get_seals(struct file *file)
2039 if (file->f_op != &shmem_file_operations)
2042 return SHMEM_I(file_inode(file))->seals;
2044 EXPORT_SYMBOL_GPL(shmem_get_seals);
2046 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2052 /* disallow upper 32bit */
2056 error = shmem_add_seals(file, arg);
2059 error = shmem_get_seals(file);
2069 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2072 struct inode *inode = file_inode(file);
2073 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2074 struct shmem_inode_info *info = SHMEM_I(inode);
2075 struct shmem_falloc shmem_falloc;
2076 pgoff_t start, index, end;
2079 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2082 mutex_lock(&inode->i_mutex);
2084 if (mode & FALLOC_FL_PUNCH_HOLE) {
2085 struct address_space *mapping = file->f_mapping;
2086 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2087 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2088 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2090 /* protected by i_mutex */
2091 if (info->seals & F_SEAL_WRITE) {
2096 shmem_falloc.waitq = &shmem_falloc_waitq;
2097 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2098 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2099 spin_lock(&inode->i_lock);
2100 inode->i_private = &shmem_falloc;
2101 spin_unlock(&inode->i_lock);
2103 if ((u64)unmap_end > (u64)unmap_start)
2104 unmap_mapping_range(mapping, unmap_start,
2105 1 + unmap_end - unmap_start, 0);
2106 shmem_truncate_range(inode, offset, offset + len - 1);
2107 /* No need to unmap again: hole-punching leaves COWed pages */
2109 spin_lock(&inode->i_lock);
2110 inode->i_private = NULL;
2111 wake_up_all(&shmem_falloc_waitq);
2112 spin_unlock(&inode->i_lock);
2117 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2118 error = inode_newsize_ok(inode, offset + len);
2122 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2127 start = offset >> PAGE_CACHE_SHIFT;
2128 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2129 /* Try to avoid a swapstorm if len is impossible to satisfy */
2130 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2135 shmem_falloc.waitq = NULL;
2136 shmem_falloc.start = start;
2137 shmem_falloc.next = start;
2138 shmem_falloc.nr_falloced = 0;
2139 shmem_falloc.nr_unswapped = 0;
2140 spin_lock(&inode->i_lock);
2141 inode->i_private = &shmem_falloc;
2142 spin_unlock(&inode->i_lock);
2144 for (index = start; index < end; index++) {
2148 * Good, the fallocate(2) manpage permits EINTR: we may have
2149 * been interrupted because we are using up too much memory.
2151 if (signal_pending(current))
2153 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2156 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
2159 /* Remove the !PageUptodate pages we added */
2160 shmem_undo_range(inode,
2161 (loff_t)start << PAGE_CACHE_SHIFT,
2162 (loff_t)index << PAGE_CACHE_SHIFT, true);
2167 * Inform shmem_writepage() how far we have reached.
2168 * No need for lock or barrier: we have the page lock.
2170 shmem_falloc.next++;
2171 if (!PageUptodate(page))
2172 shmem_falloc.nr_falloced++;
2175 * If !PageUptodate, leave it that way so that freeable pages
2176 * can be recognized if we need to rollback on error later.
2177 * But set_page_dirty so that memory pressure will swap rather
2178 * than free the pages we are allocating (and SGP_CACHE pages
2179 * might still be clean: we now need to mark those dirty too).
2181 set_page_dirty(page);
2183 page_cache_release(page);
2187 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2188 i_size_write(inode, offset + len);
2189 inode->i_ctime = CURRENT_TIME;
2191 spin_lock(&inode->i_lock);
2192 inode->i_private = NULL;
2193 spin_unlock(&inode->i_lock);
2195 mutex_unlock(&inode->i_mutex);
2199 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2201 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2203 buf->f_type = TMPFS_MAGIC;
2204 buf->f_bsize = PAGE_CACHE_SIZE;
2205 buf->f_namelen = NAME_MAX;
2206 if (sbinfo->max_blocks) {
2207 buf->f_blocks = sbinfo->max_blocks;
2209 buf->f_bfree = sbinfo->max_blocks -
2210 percpu_counter_sum(&sbinfo->used_blocks);
2212 if (sbinfo->max_inodes) {
2213 buf->f_files = sbinfo->max_inodes;
2214 buf->f_ffree = sbinfo->free_inodes;
2216 /* else leave those fields 0 like simple_statfs */
2221 * File creation. Allocate an inode, and we're done..
2224 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2226 struct inode *inode;
2227 int error = -ENOSPC;
2229 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2231 error = simple_acl_create(dir, inode);
2234 error = security_inode_init_security(inode, dir,
2236 shmem_initxattrs, NULL);
2237 if (error && error != -EOPNOTSUPP)
2241 dir->i_size += BOGO_DIRENT_SIZE;
2242 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2243 d_instantiate(dentry, inode);
2244 dget(dentry); /* Extra count - pin the dentry in core */
2253 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2255 struct inode *inode;
2256 int error = -ENOSPC;
2258 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2260 error = security_inode_init_security(inode, dir,
2262 shmem_initxattrs, NULL);
2263 if (error && error != -EOPNOTSUPP)
2265 error = simple_acl_create(dir, inode);
2268 d_tmpfile(dentry, inode);
2276 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2280 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2286 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2289 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2295 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2297 struct inode *inode = d_inode(old_dentry);
2301 * No ordinary (disk based) filesystem counts links as inodes;
2302 * but each new link needs a new dentry, pinning lowmem, and
2303 * tmpfs dentries cannot be pruned until they are unlinked.
2305 ret = shmem_reserve_inode(inode->i_sb);
2309 dir->i_size += BOGO_DIRENT_SIZE;
2310 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2312 ihold(inode); /* New dentry reference */
2313 dget(dentry); /* Extra pinning count for the created dentry */
2314 d_instantiate(dentry, inode);
2319 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2321 struct inode *inode = d_inode(dentry);
2323 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2324 shmem_free_inode(inode->i_sb);
2326 dir->i_size -= BOGO_DIRENT_SIZE;
2327 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2329 dput(dentry); /* Undo the count from "create" - this does all the work */
2333 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2335 if (!simple_empty(dentry))
2338 drop_nlink(d_inode(dentry));
2340 return shmem_unlink(dir, dentry);
2343 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2345 bool old_is_dir = d_is_dir(old_dentry);
2346 bool new_is_dir = d_is_dir(new_dentry);
2348 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2350 drop_nlink(old_dir);
2353 drop_nlink(new_dir);
2357 old_dir->i_ctime = old_dir->i_mtime =
2358 new_dir->i_ctime = new_dir->i_mtime =
2359 d_inode(old_dentry)->i_ctime =
2360 d_inode(new_dentry)->i_ctime = CURRENT_TIME;
2365 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2367 struct dentry *whiteout;
2370 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2374 error = shmem_mknod(old_dir, whiteout,
2375 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2381 * Cheat and hash the whiteout while the old dentry is still in
2382 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2384 * d_lookup() will consistently find one of them at this point,
2385 * not sure which one, but that isn't even important.
2392 * The VFS layer already does all the dentry stuff for rename,
2393 * we just have to decrement the usage count for the target if
2394 * it exists so that the VFS layer correctly free's it when it
2397 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2399 struct inode *inode = d_inode(old_dentry);
2400 int they_are_dirs = S_ISDIR(inode->i_mode);
2402 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2405 if (flags & RENAME_EXCHANGE)
2406 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2408 if (!simple_empty(new_dentry))
2411 if (flags & RENAME_WHITEOUT) {
2414 error = shmem_whiteout(old_dir, old_dentry);
2419 if (d_really_is_positive(new_dentry)) {
2420 (void) shmem_unlink(new_dir, new_dentry);
2421 if (they_are_dirs) {
2422 drop_nlink(d_inode(new_dentry));
2423 drop_nlink(old_dir);
2425 } else if (they_are_dirs) {
2426 drop_nlink(old_dir);
2430 old_dir->i_size -= BOGO_DIRENT_SIZE;
2431 new_dir->i_size += BOGO_DIRENT_SIZE;
2432 old_dir->i_ctime = old_dir->i_mtime =
2433 new_dir->i_ctime = new_dir->i_mtime =
2434 inode->i_ctime = CURRENT_TIME;
2438 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2442 struct inode *inode;
2445 struct shmem_inode_info *info;
2447 len = strlen(symname) + 1;
2448 if (len > PAGE_CACHE_SIZE)
2449 return -ENAMETOOLONG;
2451 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2455 error = security_inode_init_security(inode, dir, &dentry->d_name,
2456 shmem_initxattrs, NULL);
2458 if (error != -EOPNOTSUPP) {
2465 info = SHMEM_I(inode);
2466 inode->i_size = len-1;
2467 if (len <= SHORT_SYMLINK_LEN) {
2468 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2469 if (!info->symlink) {
2473 inode->i_op = &shmem_short_symlink_operations;
2474 inode->i_link = info->symlink;
2476 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2481 inode->i_mapping->a_ops = &shmem_aops;
2482 inode->i_op = &shmem_symlink_inode_operations;
2483 kaddr = kmap_atomic(page);
2484 memcpy(kaddr, symname, len);
2485 kunmap_atomic(kaddr);
2486 SetPageUptodate(page);
2487 set_page_dirty(page);
2489 page_cache_release(page);
2491 dir->i_size += BOGO_DIRENT_SIZE;
2492 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2493 d_instantiate(dentry, inode);
2498 static const char *shmem_follow_link(struct dentry *dentry, void **cookie)
2500 struct page *page = NULL;
2501 int error = shmem_getpage(d_inode(dentry), 0, &page, SGP_READ, NULL);
2503 return ERR_PTR(error);
2509 static void shmem_put_link(struct inode *unused, void *cookie)
2511 struct page *page = cookie;
2513 mark_page_accessed(page);
2514 page_cache_release(page);
2517 #ifdef CONFIG_TMPFS_XATTR
2519 * Superblocks without xattr inode operations may get some security.* xattr
2520 * support from the LSM "for free". As soon as we have any other xattrs
2521 * like ACLs, we also need to implement the security.* handlers at
2522 * filesystem level, though.
2526 * Callback for security_inode_init_security() for acquiring xattrs.
2528 static int shmem_initxattrs(struct inode *inode,
2529 const struct xattr *xattr_array,
2532 struct shmem_inode_info *info = SHMEM_I(inode);
2533 const struct xattr *xattr;
2534 struct simple_xattr *new_xattr;
2537 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2538 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2542 len = strlen(xattr->name) + 1;
2543 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2545 if (!new_xattr->name) {
2550 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2551 XATTR_SECURITY_PREFIX_LEN);
2552 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2555 simple_xattr_list_add(&info->xattrs, new_xattr);
2561 static const struct xattr_handler *shmem_xattr_handlers[] = {
2562 #ifdef CONFIG_TMPFS_POSIX_ACL
2563 &posix_acl_access_xattr_handler,
2564 &posix_acl_default_xattr_handler,
2569 static int shmem_xattr_validate(const char *name)
2571 struct { const char *prefix; size_t len; } arr[] = {
2572 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2573 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2577 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2578 size_t preflen = arr[i].len;
2579 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2588 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2589 void *buffer, size_t size)
2591 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2595 * If this is a request for a synthetic attribute in the system.*
2596 * namespace use the generic infrastructure to resolve a handler
2597 * for it via sb->s_xattr.
2599 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2600 return generic_getxattr(dentry, name, buffer, size);
2602 err = shmem_xattr_validate(name);
2606 return simple_xattr_get(&info->xattrs, name, buffer, size);
2609 static int shmem_setxattr(struct dentry *dentry, const char *name,
2610 const void *value, size_t size, int flags)
2612 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2616 * If this is a request for a synthetic attribute in the system.*
2617 * namespace use the generic infrastructure to resolve a handler
2618 * for it via sb->s_xattr.
2620 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2621 return generic_setxattr(dentry, name, value, size, flags);
2623 err = shmem_xattr_validate(name);
2627 return simple_xattr_set(&info->xattrs, name, value, size, flags);
2630 static int shmem_removexattr(struct dentry *dentry, const char *name)
2632 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2636 * If this is a request for a synthetic attribute in the system.*
2637 * namespace use the generic infrastructure to resolve a handler
2638 * for it via sb->s_xattr.
2640 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2641 return generic_removexattr(dentry, name);
2643 err = shmem_xattr_validate(name);
2647 return simple_xattr_remove(&info->xattrs, name);
2650 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2652 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2653 return simple_xattr_list(&info->xattrs, buffer, size);
2655 #endif /* CONFIG_TMPFS_XATTR */
2657 static const struct inode_operations shmem_short_symlink_operations = {
2658 .readlink = generic_readlink,
2659 .follow_link = simple_follow_link,
2660 #ifdef CONFIG_TMPFS_XATTR
2661 .setxattr = shmem_setxattr,
2662 .getxattr = shmem_getxattr,
2663 .listxattr = shmem_listxattr,
2664 .removexattr = shmem_removexattr,
2668 static const struct inode_operations shmem_symlink_inode_operations = {
2669 .readlink = generic_readlink,
2670 .follow_link = shmem_follow_link,
2671 .put_link = shmem_put_link,
2672 #ifdef CONFIG_TMPFS_XATTR
2673 .setxattr = shmem_setxattr,
2674 .getxattr = shmem_getxattr,
2675 .listxattr = shmem_listxattr,
2676 .removexattr = shmem_removexattr,
2680 static struct dentry *shmem_get_parent(struct dentry *child)
2682 return ERR_PTR(-ESTALE);
2685 static int shmem_match(struct inode *ino, void *vfh)
2689 inum = (inum << 32) | fh[1];
2690 return ino->i_ino == inum && fh[0] == ino->i_generation;
2693 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2694 struct fid *fid, int fh_len, int fh_type)
2696 struct inode *inode;
2697 struct dentry *dentry = NULL;
2704 inum = (inum << 32) | fid->raw[1];
2706 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2707 shmem_match, fid->raw);
2709 dentry = d_find_alias(inode);
2716 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2717 struct inode *parent)
2721 return FILEID_INVALID;
2724 if (inode_unhashed(inode)) {
2725 /* Unfortunately insert_inode_hash is not idempotent,
2726 * so as we hash inodes here rather than at creation
2727 * time, we need a lock to ensure we only try
2730 static DEFINE_SPINLOCK(lock);
2732 if (inode_unhashed(inode))
2733 __insert_inode_hash(inode,
2734 inode->i_ino + inode->i_generation);
2738 fh[0] = inode->i_generation;
2739 fh[1] = inode->i_ino;
2740 fh[2] = ((__u64)inode->i_ino) >> 32;
2746 static const struct export_operations shmem_export_ops = {
2747 .get_parent = shmem_get_parent,
2748 .encode_fh = shmem_encode_fh,
2749 .fh_to_dentry = shmem_fh_to_dentry,
2752 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2755 char *this_char, *value, *rest;
2756 struct mempolicy *mpol = NULL;
2760 while (options != NULL) {
2761 this_char = options;
2764 * NUL-terminate this option: unfortunately,
2765 * mount options form a comma-separated list,
2766 * but mpol's nodelist may also contain commas.
2768 options = strchr(options, ',');
2769 if (options == NULL)
2772 if (!isdigit(*options)) {
2779 if ((value = strchr(this_char,'=')) != NULL) {
2783 "tmpfs: No value for mount option '%s'\n",
2788 if (!strcmp(this_char,"size")) {
2789 unsigned long long size;
2790 size = memparse(value,&rest);
2792 size <<= PAGE_SHIFT;
2793 size *= totalram_pages;
2799 sbinfo->max_blocks =
2800 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2801 } else if (!strcmp(this_char,"nr_blocks")) {
2802 sbinfo->max_blocks = memparse(value, &rest);
2805 } else if (!strcmp(this_char,"nr_inodes")) {
2806 sbinfo->max_inodes = memparse(value, &rest);
2809 } else if (!strcmp(this_char,"mode")) {
2812 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2815 } else if (!strcmp(this_char,"uid")) {
2818 uid = simple_strtoul(value, &rest, 0);
2821 sbinfo->uid = make_kuid(current_user_ns(), uid);
2822 if (!uid_valid(sbinfo->uid))
2824 } else if (!strcmp(this_char,"gid")) {
2827 gid = simple_strtoul(value, &rest, 0);
2830 sbinfo->gid = make_kgid(current_user_ns(), gid);
2831 if (!gid_valid(sbinfo->gid))
2833 } else if (!strcmp(this_char,"mpol")) {
2836 if (mpol_parse_str(value, &mpol))
2839 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2844 sbinfo->mpol = mpol;
2848 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2856 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2858 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2859 struct shmem_sb_info config = *sbinfo;
2860 unsigned long inodes;
2861 int error = -EINVAL;
2864 if (shmem_parse_options(data, &config, true))
2867 spin_lock(&sbinfo->stat_lock);
2868 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2869 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2871 if (config.max_inodes < inodes)
2874 * Those tests disallow limited->unlimited while any are in use;
2875 * but we must separately disallow unlimited->limited, because
2876 * in that case we have no record of how much is already in use.
2878 if (config.max_blocks && !sbinfo->max_blocks)
2880 if (config.max_inodes && !sbinfo->max_inodes)
2884 sbinfo->max_blocks = config.max_blocks;
2885 sbinfo->max_inodes = config.max_inodes;
2886 sbinfo->free_inodes = config.max_inodes - inodes;
2889 * Preserve previous mempolicy unless mpol remount option was specified.
2892 mpol_put(sbinfo->mpol);
2893 sbinfo->mpol = config.mpol; /* transfers initial ref */
2896 spin_unlock(&sbinfo->stat_lock);
2900 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2902 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2904 if (sbinfo->max_blocks != shmem_default_max_blocks())
2905 seq_printf(seq, ",size=%luk",
2906 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2907 if (sbinfo->max_inodes != shmem_default_max_inodes())
2908 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2909 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2910 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2911 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2912 seq_printf(seq, ",uid=%u",
2913 from_kuid_munged(&init_user_ns, sbinfo->uid));
2914 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2915 seq_printf(seq, ",gid=%u",
2916 from_kgid_munged(&init_user_ns, sbinfo->gid));
2917 shmem_show_mpol(seq, sbinfo->mpol);
2921 #define MFD_NAME_PREFIX "memfd:"
2922 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2923 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2925 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2927 SYSCALL_DEFINE2(memfd_create,
2928 const char __user *, uname,
2929 unsigned int, flags)
2931 struct shmem_inode_info *info;
2937 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2940 /* length includes terminating zero */
2941 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2944 if (len > MFD_NAME_MAX_LEN + 1)
2947 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2951 strcpy(name, MFD_NAME_PREFIX);
2952 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2957 /* terminating-zero may have changed after strnlen_user() returned */
2958 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
2963 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
2969 file = shmem_file_setup(name, 0, VM_NORESERVE);
2971 error = PTR_ERR(file);
2974 info = SHMEM_I(file_inode(file));
2975 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
2976 file->f_flags |= O_RDWR | O_LARGEFILE;
2977 if (flags & MFD_ALLOW_SEALING)
2978 info->seals &= ~F_SEAL_SEAL;
2980 fd_install(fd, file);
2991 #endif /* CONFIG_TMPFS */
2993 static void shmem_put_super(struct super_block *sb)
2995 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2997 percpu_counter_destroy(&sbinfo->used_blocks);
2998 mpol_put(sbinfo->mpol);
3000 sb->s_fs_info = NULL;
3003 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3005 struct inode *inode;
3006 struct shmem_sb_info *sbinfo;
3009 /* Round up to L1_CACHE_BYTES to resist false sharing */
3010 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3011 L1_CACHE_BYTES), GFP_KERNEL);
3015 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3016 sbinfo->uid = current_fsuid();
3017 sbinfo->gid = current_fsgid();
3018 sb->s_fs_info = sbinfo;
3022 * Per default we only allow half of the physical ram per
3023 * tmpfs instance, limiting inodes to one per page of lowmem;
3024 * but the internal instance is left unlimited.
3026 if (!(sb->s_flags & MS_KERNMOUNT)) {
3027 sbinfo->max_blocks = shmem_default_max_blocks();
3028 sbinfo->max_inodes = shmem_default_max_inodes();
3029 if (shmem_parse_options(data, sbinfo, false)) {
3034 sb->s_flags |= MS_NOUSER;
3036 sb->s_export_op = &shmem_export_ops;
3037 sb->s_flags |= MS_NOSEC;
3039 sb->s_flags |= MS_NOUSER;
3042 spin_lock_init(&sbinfo->stat_lock);
3043 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3045 sbinfo->free_inodes = sbinfo->max_inodes;
3047 sb->s_maxbytes = MAX_LFS_FILESIZE;
3048 sb->s_blocksize = PAGE_CACHE_SIZE;
3049 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3050 sb->s_magic = TMPFS_MAGIC;
3051 sb->s_op = &shmem_ops;
3052 sb->s_time_gran = 1;
3053 #ifdef CONFIG_TMPFS_XATTR
3054 sb->s_xattr = shmem_xattr_handlers;
3056 #ifdef CONFIG_TMPFS_POSIX_ACL
3057 sb->s_flags |= MS_POSIXACL;
3060 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3063 inode->i_uid = sbinfo->uid;
3064 inode->i_gid = sbinfo->gid;
3065 sb->s_root = d_make_root(inode);
3071 shmem_put_super(sb);
3075 static struct kmem_cache *shmem_inode_cachep;
3077 static struct inode *shmem_alloc_inode(struct super_block *sb)
3079 struct shmem_inode_info *info;
3080 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3083 return &info->vfs_inode;
3086 static void shmem_destroy_callback(struct rcu_head *head)
3088 struct inode *inode = container_of(head, struct inode, i_rcu);
3089 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3092 static void shmem_destroy_inode(struct inode *inode)
3094 if (S_ISREG(inode->i_mode))
3095 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3096 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3099 static void shmem_init_inode(void *foo)
3101 struct shmem_inode_info *info = foo;
3102 inode_init_once(&info->vfs_inode);
3105 static int shmem_init_inodecache(void)
3107 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3108 sizeof(struct shmem_inode_info),
3109 0, SLAB_PANIC, shmem_init_inode);
3113 static void shmem_destroy_inodecache(void)
3115 kmem_cache_destroy(shmem_inode_cachep);
3118 static const struct address_space_operations shmem_aops = {
3119 .writepage = shmem_writepage,
3120 .set_page_dirty = __set_page_dirty_no_writeback,
3122 .write_begin = shmem_write_begin,
3123 .write_end = shmem_write_end,
3125 #ifdef CONFIG_MIGRATION
3126 .migratepage = migrate_page,
3128 .error_remove_page = generic_error_remove_page,
3131 static const struct file_operations shmem_file_operations = {
3134 .llseek = shmem_file_llseek,
3135 .read_iter = shmem_file_read_iter,
3136 .write_iter = generic_file_write_iter,
3137 .fsync = noop_fsync,
3138 .splice_read = shmem_file_splice_read,
3139 .splice_write = iter_file_splice_write,
3140 .fallocate = shmem_fallocate,
3144 static const struct inode_operations shmem_inode_operations = {
3145 .getattr = shmem_getattr,
3146 .setattr = shmem_setattr,
3147 #ifdef CONFIG_TMPFS_XATTR
3148 .setxattr = shmem_setxattr,
3149 .getxattr = shmem_getxattr,
3150 .listxattr = shmem_listxattr,
3151 .removexattr = shmem_removexattr,
3152 .set_acl = simple_set_acl,
3156 static const struct inode_operations shmem_dir_inode_operations = {
3158 .create = shmem_create,
3159 .lookup = simple_lookup,
3161 .unlink = shmem_unlink,
3162 .symlink = shmem_symlink,
3163 .mkdir = shmem_mkdir,
3164 .rmdir = shmem_rmdir,
3165 .mknod = shmem_mknod,
3166 .rename2 = shmem_rename2,
3167 .tmpfile = shmem_tmpfile,
3169 #ifdef CONFIG_TMPFS_XATTR
3170 .setxattr = shmem_setxattr,
3171 .getxattr = shmem_getxattr,
3172 .listxattr = shmem_listxattr,
3173 .removexattr = shmem_removexattr,
3175 #ifdef CONFIG_TMPFS_POSIX_ACL
3176 .setattr = shmem_setattr,
3177 .set_acl = simple_set_acl,
3181 static const struct inode_operations shmem_special_inode_operations = {
3182 #ifdef CONFIG_TMPFS_XATTR
3183 .setxattr = shmem_setxattr,
3184 .getxattr = shmem_getxattr,
3185 .listxattr = shmem_listxattr,
3186 .removexattr = shmem_removexattr,
3188 #ifdef CONFIG_TMPFS_POSIX_ACL
3189 .setattr = shmem_setattr,
3190 .set_acl = simple_set_acl,
3194 static const struct super_operations shmem_ops = {
3195 .alloc_inode = shmem_alloc_inode,
3196 .destroy_inode = shmem_destroy_inode,
3198 .statfs = shmem_statfs,
3199 .remount_fs = shmem_remount_fs,
3200 .show_options = shmem_show_options,
3202 .evict_inode = shmem_evict_inode,
3203 .drop_inode = generic_delete_inode,
3204 .put_super = shmem_put_super,
3207 static const struct vm_operations_struct shmem_vm_ops = {
3208 .fault = shmem_fault,
3209 .map_pages = filemap_map_pages,
3211 .set_policy = shmem_set_policy,
3212 .get_policy = shmem_get_policy,
3216 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3217 int flags, const char *dev_name, void *data)
3219 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3222 static struct file_system_type shmem_fs_type = {
3223 .owner = THIS_MODULE,
3225 .mount = shmem_mount,
3226 .kill_sb = kill_litter_super,
3227 .fs_flags = FS_USERNS_MOUNT,
3230 int __init shmem_init(void)
3234 /* If rootfs called this, don't re-init */
3235 if (shmem_inode_cachep)
3238 error = shmem_init_inodecache();
3242 error = register_filesystem(&shmem_fs_type);
3244 printk(KERN_ERR "Could not register tmpfs\n");
3248 shm_mnt = kern_mount(&shmem_fs_type);
3249 if (IS_ERR(shm_mnt)) {
3250 error = PTR_ERR(shm_mnt);
3251 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3257 unregister_filesystem(&shmem_fs_type);
3259 shmem_destroy_inodecache();
3261 shm_mnt = ERR_PTR(error);
3265 #else /* !CONFIG_SHMEM */
3268 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3270 * This is intended for small system where the benefits of the full
3271 * shmem code (swap-backed and resource-limited) are outweighed by
3272 * their complexity. On systems without swap this code should be
3273 * effectively equivalent, but much lighter weight.
3276 static struct file_system_type shmem_fs_type = {
3278 .mount = ramfs_mount,
3279 .kill_sb = kill_litter_super,
3280 .fs_flags = FS_USERNS_MOUNT,
3283 int __init shmem_init(void)
3285 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3287 shm_mnt = kern_mount(&shmem_fs_type);
3288 BUG_ON(IS_ERR(shm_mnt));
3293 int shmem_unuse(swp_entry_t swap, struct page *page)
3298 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3303 void shmem_unlock_mapping(struct address_space *mapping)
3307 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3309 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3311 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3313 #define shmem_vm_ops generic_file_vm_ops
3314 #define shmem_file_operations ramfs_file_operations
3315 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3316 #define shmem_acct_size(flags, size) 0
3317 #define shmem_unacct_size(flags, size) do {} while (0)
3319 #endif /* CONFIG_SHMEM */
3323 static struct dentry_operations anon_ops = {
3324 .d_dname = simple_dname
3327 static struct file *__shmem_file_setup(const char *name, loff_t size,
3328 unsigned long flags, unsigned int i_flags)
3331 struct inode *inode;
3333 struct super_block *sb;
3336 if (IS_ERR(shm_mnt))
3337 return ERR_CAST(shm_mnt);
3339 if (size < 0 || size > MAX_LFS_FILESIZE)
3340 return ERR_PTR(-EINVAL);
3342 if (shmem_acct_size(flags, size))
3343 return ERR_PTR(-ENOMEM);
3345 res = ERR_PTR(-ENOMEM);
3347 this.len = strlen(name);
3348 this.hash = 0; /* will go */
3349 sb = shm_mnt->mnt_sb;
3350 path.mnt = mntget(shm_mnt);
3351 path.dentry = d_alloc_pseudo(sb, &this);
3354 d_set_d_op(path.dentry, &anon_ops);
3356 res = ERR_PTR(-ENOSPC);
3357 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3361 inode->i_flags |= i_flags;
3362 d_instantiate(path.dentry, inode);
3363 inode->i_size = size;
3364 clear_nlink(inode); /* It is unlinked */
3365 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3369 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3370 &shmem_file_operations);
3377 shmem_unacct_size(flags, size);
3384 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3385 * kernel internal. There will be NO LSM permission checks against the
3386 * underlying inode. So users of this interface must do LSM checks at a
3387 * higher layer. The users are the big_key and shm implementations. LSM
3388 * checks are provided at the key or shm level rather than the inode.
3389 * @name: name for dentry (to be seen in /proc/<pid>/maps
3390 * @size: size to be set for the file
3391 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3393 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3395 return __shmem_file_setup(name, size, flags, S_PRIVATE);
3399 * shmem_file_setup - get an unlinked file living in tmpfs
3400 * @name: name for dentry (to be seen in /proc/<pid>/maps
3401 * @size: size to be set for the file
3402 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3404 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3406 return __shmem_file_setup(name, size, flags, 0);
3408 EXPORT_SYMBOL_GPL(shmem_file_setup);
3411 * shmem_zero_setup - setup a shared anonymous mapping
3412 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3414 int shmem_zero_setup(struct vm_area_struct *vma)
3417 loff_t size = vma->vm_end - vma->vm_start;
3420 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3421 * between XFS directory reading and selinux: since this file is only
3422 * accessible to the user through its mapping, use S_PRIVATE flag to
3423 * bypass file security, in the same way as shmem_kernel_file_setup().
3425 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
3427 return PTR_ERR(file);
3431 vma->vm_file = file;
3432 vma->vm_ops = &shmem_vm_ops;
3437 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3438 * @mapping: the page's address_space
3439 * @index: the page index
3440 * @gfp: the page allocator flags to use if allocating
3442 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3443 * with any new page allocations done using the specified allocation flags.
3444 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3445 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3446 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3448 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3449 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3451 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3452 pgoff_t index, gfp_t gfp)
3455 struct inode *inode = mapping->host;
3459 BUG_ON(mapping->a_ops != &shmem_aops);
3460 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3462 page = ERR_PTR(error);
3468 * The tiny !SHMEM case uses ramfs without swap
3470 return read_cache_page_gfp(mapping, index, gfp);
3473 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);